Standard Measurements

Suppose you've detected nuclei and want to know how big each one is and how much Ki67 it holds. Standard Measurements answers both, from two inputs. Coded is the coded imageLoading... of events, and it supplies every shape measurement. Shades is a single grayscale imageLoading..., and it supplies every intensity measurement.

Shades can be a marker channel or a virtual channel made by another engine, such as a distance transform used to find the closest structure. In the reference example it is a Ki67 image from Color SeparationLoading..., measured over detected structures in an IHC colon sample. The global version adds a third input, Mask, which limits the measurements to its white areas.

The engine comes in two versions. The global version, under Add Measurements (basic), measures global events: pieces of one structure that touch across neighboring fields of view, stitched into one. It needs Stitch objects enabled in the layer's options, which is the default. The local version, under Add Measurements (advanced), measures each field of view's events separately and needs Stitch objects disabled.

What you feed Shades determines what an intensity measurement means. With a distance mapLoading... as Shades, an intensity becomes a distance. With the ClassifierLoading...'s Measurements Gray Map as Shades, Most Frequent Intensity gives each nucleus the class covering most of it.

Shape descriptors. In image analysis generally, shape descriptors are ratios that don't depend on size. The textbook compactness, also called circularity, is 4πA/P², which is 1 for a perfect circle and falls as the outline grows irregular. The Engines Reference gives no formula, only the behavior: close to 1 for a disk, closer to 0 otherwise. Perimeter is hard to measure on a pixel grid, so in general small objects give noisy compactness values.

Eccentricity and its relatives. The textbook eccentricity is that of the ellipse with the same second moments as the object: 0 for a circle, approaching 1 for a line. StrataQuest's description runs over the same range. The Feret ratio, minimum caliper size over maximum, runs the other way, from 1 for a circle toward 0 for a line. Equivalent diameter, the diameter of a circle with the object's area, measures size and ignores shape.

Choosing an intensity statistic. Mean suits a marker spread evenly through the object. Sum grows with object size, so it tracks total content rather than concentration. For a marker gathered in a few bright granules, Percentile and Upper mean (the average of the pixels above the percentile) catch the bright fraction that a mean dilutes.

Intensity is not concentration. In fluorescence imaging generally, the intensity of a pixel depends on more than the amount of fluorophore. Uneven illumination (vignettingLoading...), focus, photobleaching and quenching all change it. Comparing absolute intensities between cells or slides holds only when those factors are controlled or corrected. Averages over many cells are steadier than single-cell values, because independent errors partly cancel.

Global version. Shape: Area, Perimeter, Compactness, Centroid -X, Centroid -Y, Equivalent diameter, Curvature (v1) (faster, less accurate), Curvature (v2) (slower, more accurate), Eccentricity, Minimum Width, Maximum Length, Feret Ratio and Orientation (the angle of the maximum length from horizontal, counterclockwise). Intensity: Mean, Minimum, Maximum, Range, Sum, Variance, Standard deviation, Percentile, Lower mean, Upper mean, Most Frequent Intensity, Percent of non-zero and Most frequent value.

Local version. Shape: Area, Perimeter, Compactness, Equivalent diameter, Eccentricity, Minimum caliper diameter, Maximum caliper diameter and Feret ratio. Intensity: Mean through Upper mean, as in the global list, without the last three.

Lower mean and Upper mean are the averages of the pixels below and above the chosen Percentile. The Default selection is a reasonable start; add what your question needs, since each measurement becomes a column in Raw DataLoading... and an axis you can plot.

Source: StrataQuest 8.1 Engines Reference, ch. 41 (pp. 537–543), ch. 8 (p. 375), ch. 7 (Cellular, pp. 361–362), ch. 24 (p. 438); User Manual (pp. 151, 154).

Here is how a 6-marker multiplex IF panel could be measured.

1. Detect nuclei on DAPI, and build a ring around each nucleus with the Cellular engine's ring mask or Grow (global)Loading... in Ring mode.
2. Add one Standard Measurements engine per marker on the nuclei, with Shades set to DAPI, CD3, CD8, CD20, PD-L1 and CK in turn. Keep the Default selection, and add Compactness on the DAPI one.
3. Add six more on the ring mask, for the same markers outside the nucleus.
4. Plot the means on scattergramsLoading... and gate T cells (CD3+), cytotoxic T cells (CD3+CD8+), B cells (CD20+) and tumor cells (CK+) as phenotypesLoading....

That makes twelve engines for six markers in two compartments. The "Add Single cell analysis (with cell masks) on" quick action builds a similar set for you: Nuclei detection with Standard measurements, and for every other channel a cell mask with its own Standard measurement.